| Bus,metro and taxi as the city's main public transport,carrying the daily travel of urban residents.The trajectory data recorded by each vehicle and the information of the smart card containing rich residents travel behavior information,and the different modes of residents' travel can be summarized through the data mining.The most commonly used method for resident travel pattern mining is clustering.However,the traditional clustering method is not suitable for the geographical data,which has abundant spatiotemporal information.Therefore,we need to carry out space-time expansion of traditional methods to meet the research needs.In this paper,we take the taxi trajectory data and smart card data of Shenzhen as the experimental data set.Counting the number and percentage of the residents' travel using three different modes of transport in different areas,exploring and analyzing resident trip behavior with three type of transportation in different areas.And then,using the space-time clustering method based on adjacency to excavate the specific residents 'travel mode,and understanding the residents' travel mode from the three aspects of transportation relationship,social background and OD flows.The main work of this paper are as follows:(1)Spatiotemporal expansion of spectral clustering methodThe spectral clustering method is based on graph theory,it can cluster any shape of the sample space,and the results can converge to global optimization.In this paper,the principle of spectral clustering method and the discretization method are summarized.We pointed out that the key step of spectral clustering algorithm is the construction of similarity matrix.By analyzing the similarity matrix of traditional spectral clustering,it is found that only use the European space distance to calculate the similarity between the time and space trajectories is very limited.Therefore,we propose a construction method of time-space similarity matrix considering both time series similarity and spatial adjacency distance,so as to realize the spatiotemporal expansion of spectral clustering.(2)Spatiotemporal clustering of multi-source trajectory dataMulti-source trajectory data from different public transport,recorded a wealth of time and space information and residents travel rules,if the multi-source trajectory data can be mixed to use,with the time-space clustering method,we can get more valuable potential information according to cluster results.In view of the mutual independence of multi-source trajectory data,it is necessary to establish the same metric according to the actual usage of the traffic in the residents.The ratio of travel usage of each vehicle to the total travel volume is calculated,and the information of multi-source trajectory data is classified into the unified proportional measurement system,and then the spatial and temporal similarity matrix under each traffic condition is calculated.Finally,Domain image imaging principle to realize the fusion of spatiotemporal similarity matrices of multi-source trajectory data,so as to complete the space-time clustering of multi-source trajectory data and lay the foundation for the mining of residents' travel mode.(3)Mining and Analysis of Resident Trip ModeSpatiotemporal clustering of multi-source trajectory data is carried out by using space-time clustering method based on adjacency distance.Through the analysis and mining of clustering results,5 daily travel patterns of residents were found.In order to understand the characteristics of travel mode better,we analyzed the transportation relationship,social background and OD flows of each mode.We found the operational competition and cooperative cooperation between bus,subway and taxi are different in different regions and residents 'travel mode,and the difference of the choice of residents' travels in different social background areas is combined with the travel situation,Indicating the travel patterns in different spatial regions of the interactive relationship. |
PDF Full Text Request